wisp-science

Wisp Science (Workspace for Intelligent Scientific Practice) is a desktop AI research assistant and scientific computing workbench, also available as a headless CLI. It is built with Rust, Tauri v2 and Leptos. It connects to OpenAI-compatible and Anthropic models, runs persistent Python and R environments on local, WSL, SSH and GPU machines, loads reusable Agent Skills (`SKILL.md`), and reaches roughly 80 bioinformatics and computational biology databases such as PubMed and GEO through bundled MCP servers. Your data, conversations and credentials stay on your own machines.

The agent can read, write, search and run shell commands inside a project-rooted path sandbox behind explicit approval gates, with an opt-in Full Permission mode per conversation. Long jobs run as structured Runs with preflight checks, heartbeats and log tails. Secrets are kept in the OS keyring, and SSH hosts are registered and probed rather than used through free-form commands. Other features include offline previews of notebooks, PDF and Office files, local Markdown extraction, driving external coding agents such as Codex and Claude Code over ACP, sub-agent teams with Controlled Delegation, and a Publication Workspace that exports deterministic Evidence Capsules.

Conversations persist in SQLite, a turn's file edits can be undone with a preview, and sync is manual and encrypted rather than in the background. Installers are on GitHub Releases for Windows (unsigned, so SmartScreen warns) and macOS (unsigned, Apple Silicon and Intel), with a Rust and Tauri build from source. The README labels the project an MVP vertical slice with a post-MVP roadmap. The licence is AGPL-3.0, which requires you to share source if you offer a modified version as a network service. Last push was 2026-10-13. This review is based on the README and repository data; we have not run it ourselves.